Toward a robust detection of viscous and turbulent flow regions using unsupervised machine learning

We propose an invariant feature space for the detection of viscous dominated and turbulent regions (i.e., boundary layers wakes). The developed methodology uses principal invariants strain rotational rate tensors as input to unsupervised Machine Learning Gaussian mixture model. selected is independent coordinate frame used generate processed data, it relies on rate, which are Galilean invariants. This allows us identify two distinct flow regions: a dominated, region (boundary layer wake region) inviscid, irrotational (outer region). test laminar (using Large Eddy Simulation) case flows past circular cylinder at $Re=40$ $Re=3900$. simulations have been conducted using high-order nodal Discontinuous Galerkin Spectral Element Method (DGSEM). results obtained analysed show that clustering provides effective identification method in flow. also include comparisons with traditional sensors proposed does not depend selection arbitrary threshold, required when sensors.